JPWO2013093994A1 - Storage system, data rebalancing program, and data rebalancing method - Google Patents

Abstract

The storage system stores a plurality of storage devices that store data, a cache memory that stores data, and an information processing terminal that receives an access request for reading or writing target data. An access control unit for accessing the device and storing the target data in the cache memory; a writing unit for writing the target data stored in the cache memory to a storage device in which the target data is not stored among the plurality of storage devices; including.

Description

  The technology described in this specification relates to a storage system.

  The storage system includes a plurality of storage devices and a server device that controls them. In a scale-out storage system, data rebalancing (data rearrangement) processing is performed between storage devices in order to ensure data leveling when adding storage devices. In many cases, data rebalancing copies uneven data in a storage device to another storage device. Therefore, network bandwidth and system resources are used during data rebalancing.

  Regarding the logical volume load distribution, there are the following logical volume load distribution techniques. In the storage apparatus, the performance measurement mechanism measures the load status of the logical volume based on the data amount transferred by the data transfer mechanism and the command processing amount. The copy mechanism copies the contents of the logical volume set in the physical volume to the logical volume set in the spare physical volume based on the measurement result of the performance measurement mechanism. By having multiple copied data, load distribution of data access is performed.

JP 2006-053601 A JP 2007-149068 A JP 2005-284632 A JP 2006-99748 A JP 2004-5634 A JP 2007-72538 A

  In normal data rebalancing, data stored in the storage device can be copied to the added storage, thereby avoiding concentration of access to a specific storage. This copy processing uses a network bandwidth between storage devices. Therefore, copying data between storage devices places a load on the network, so it is conceivable that copying is performed during a time zone when the load on the storage device is low.

  However, in the case of a storage device that has integrated storage, since the storage device is used by a plurality of systems, there may be a case where it is not possible to secure a time zone during which the network has a low load. In addition, since the network bandwidth is used at a time during data copying, the copying becomes a factor of input / output bottleneck between the storage apparatus and the server.

  In one aspect, the present invention provides a technique for improving the efficiency of data rebalance in a storage system.

  The storage system includes a plurality of storage devices, a cache memory, an access control unit, and a writing unit. The storage device stores data. The cache memory holds data. When there is an access request for reading target data or writing target data from the information processing terminal, the access control unit accesses any of the storage devices and stores the target data in the cache memory . The writing unit writes the target data stored in the cache memory to a storage device in which the target data is not stored among the plurality of storage devices.

  According to one embodiment, the efficiency of data rebalance in a storage system can be improved.

The structure of the storage system in this embodiment is shown. 1 is an overall configuration diagram of a system in the present embodiment. A block diagram of the volume management server 15 and the I / O server 18 is shown. An example of the replica table 31 is shown. An example of the write hold table 32 is shown. An example of the I / O periodic table 33 is shown. An example of the read / write access load correspondence table 34 is shown. An example of the read / write disk access load correspondence table 35 is shown. An example of the replica completeness table 36 is shown. An example of the disk I / O periodic table 37 is shown. An example of the read / write position access load correspondence table 38 is shown. An example of the storage capacity table 39 is shown. An example of the load statistical information 40 is shown. The data reading flow in this embodiment is shown. The data write flow in this embodiment is shown. The replica creation processing flow (the 1) in this embodiment is shown. The replica creation processing flow (No. 2) in this embodiment is shown. The replica deletion processing flow (the 1) in this embodiment is shown. The replica deletion processing flow (the 2) in this embodiment is shown. The process flow of creating a predicted replica using the I / O periodic table is shown. A work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the storage device A are tabulated every hour and a graph (B) thereof are shown. A work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the storage device B are tabulated every hour and a graph (B) thereof are shown. A work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the storage device C are counted every hour, and a graph (B) thereof are shown. The average value of the access frequency for every hour of the work table (A) shown in FIGS. 21 to 23 and its graph (B) are shown. A work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the disk B1 of the storage apparatus B are counted every minute and a graph (B) thereof are shown. . A work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the disk B2 of the storage device B are counted every minute and a graph (B) thereof are shown. . A work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the disk B3 of the storage device B are counted every minute and a graph (B) thereof are shown. . The average value of the access frequency per minute of the work table (A) shown in FIGS. 25 to 27 and the graph (B) are shown.

  In the logical volume load distribution technique described above, load distribution is performed by analyzing the access frequency and independently creating a replica and a copy of some data. However, in this case, a bandwidth for copying is required in addition to the input / output (I / O) used in the business, and this copy may affect the business. Further, the performance of the logical volume can be enhanced only at the initial setting for defining the logical volume, and the on-demand performance enhancement cannot be performed. Further, when a plurality of mirrors are created and writing is performed on the mirrors, a method of disassembling the mirrors is used when the load due to writing increases. However, there is also a problem that disassembling the mirror due to a temporary load results in performance degradation. Furthermore, since the network bandwidth is used when reading data, the network bandwidth between the storage apparatus and the client is compressed. Therefore, the above-described logical volume load distribution technique cannot be used in a large-scale storage system.

  Therefore, the efficiency of data rebalancing that is performed when a storage device is added or when the load is not balanced during operation is improved. At this time, when replicating data to a plurality of storage devices, the access request from the information processing terminal is efficiently used to improve the input / output performance of the storage system.

  FIG. 1 shows the configuration of a storage system in this embodiment. The storage system 1 in this embodiment includes a plurality of storage devices 2, a cache memory 3, an access control unit 4, and a writing unit 5.

  The storage device 2 stores data. An example of the storage device 2 is the storage device 25. The cache memory 3 holds data. An example of the cache memory 3 is an example of the cache memory 18d.

  When there is an access request for reading target data or writing target data from the information processing terminal 9, the access control unit 4 accesses one of the storage devices 2 and stores the target data in the cache memory 3. To do. An example of the access control unit 4 is an I / O server 18.

  The writing unit 5 writes the target data stored in the cache memory 3 to the storage device 2 in which the target data is not stored among the plurality of storage devices 2. An example of the writing unit 5 is an I / O server 18.

  With this configuration, it is possible to improve the efficiency of rebalancing processing between storages. That is, when a storage device is added, by effectively using an access request from the information processing terminal 9, rebalancing processing between storages can be performed efficiently without imposing a load on the network.

  The storage system further includes a load monitoring unit 6. The load monitoring unit 6 monitors the access load by reading or writing on the storage device 2. An example of the load monitoring unit 6 is an example of the load monitoring unit 17.

  When there is an access request from the information processing terminal 9, the access control unit 4 accesses the storage device with the smallest access load among the storage devices 2 and stores the target data in the cache memory 3 based on the monitoring result. To do.

  When there is a read request to the target data from the information processing terminal 9, the access control unit 4 retrieves the target data from the storage device with the least access load among the storage devices 2 that store the target data based on the monitoring result. get. The access control unit 4 stores the target data in the cache memory 3 and transmits the target data to the information processing terminal.

  When there is a request for writing the target data from the information processing terminal 9, the access control unit 4 writes the target data to the storage device with the least access load among the storage devices 2 based on the monitoring result, and caches the target data. Store in memory 3.

  The writing unit 5 writes the target data stored in the cache memory 3 to a storage device 2 other than the storage device 2 that performed the writing.

  With this configuration, the target data can be written to the storage device with the least access load among the storage devices 2 based on the monitoring result.

  The storage system 1 further includes a replica deletion unit 7. When the write access load of the replica of the virtual volume obtained by virtualizing the volumes of the plurality of storage devices 2 exceeds the threshold, the replica deletion unit 7 has any of three or more virtual volume replicas in the plurality of storage devices 2. The replica stored in the storage device 2 is deleted. Alternatively, if the free capacity of the storage device 2 is less than the first threshold value when creating a replica, the replica deletion unit 7 determines that any one of the storage devices 2 has three or more virtual volume replicas. Delete stored replicas.

  With this configuration, when there are frequent write requests from the information processing terminal or when the free capacity of the storage device becomes low, one of the virtual volumes having three or more replicas is deleted. be able to.

  The replica deletion unit 7 selects a replica of the storage device 2 with the least access load from the storage devices 2 including the replica based on the monitoring result. Based on the access frequency information indicating the access frequency with respect to the storage device 2 aggregated for each time, the replica deletion unit 7 has no read access with the access frequency exceeding the second threshold after the current time on the selected replica. If it is determined, delete the selected replica.

  With this configuration, when access exceeding the threshold value continues for a certain time after writing, the replica of the virtual volume can be excluded from the deletion target.

  The storage system 1 further includes a replica creation unit 8. The replica creation unit 8 determines that the difference between the access load in any time zone among the access loads for each time for any storage device 2 and the average of the access loads in the time zone for all storage devices 2 is the third. If the threshold is exceeded, the following processing is performed. That is, the replica creation unit 8 subdivides the time zone. The replica creation unit 8 accesses the access load on any disk of any storage device 2 in any subdivided time zone and any subdivided time of all disks in any storage device 2 It is determined whether the difference from the average access load of the band is smaller than the fourth threshold. When the difference is smaller than the fourth threshold value, the replica creating unit 8 is the most before the time when the access load of any storage device does not exceed the fifth threshold value before any of the subdivided time zones. Create a replica on a disk with a high access load.

  With this configuration, it is possible to create a replica on a disk with the highest access load for a storage apparatus that has a low access load on each disk but is heavily loaded as a whole storage apparatus.

  The storage system in this embodiment has two or more storage devices. The following description is based on the assumption that there are two storage apparatuses A and B. It is assumed that the disk in the storage apparatus A maintains a mirror state (that is, two replicas) with the disk in the storage apparatus B. These two disks are shown as one virtual disk in the I / O server.

  Here, the reason for maintaining the mirror state is that even if a disk or storage device fails, data lost or operation is not stopped. Assume that there are a plurality of client devices that access the storage as in the cloud system.

  FIG. 2 shows an overall configuration diagram of a system in the present embodiment. In the system according to this embodiment, a client device 11 and a storage system 14 are connected by a LAN (Local Area Network) 13. The client device 11 is connected to the LAN 13 via a communication interface (hereinafter, the interface is referred to as “IF”) 12. The LAN 13 is an example, and may be the Internet, an intranet, or another network.

  The storage system 14 includes a volume management server 15, an I / O server 18, a storage device 25, a management LAN 27, and an I / O LAN 28.

  The volume management server 15 controls the entire storage system 14 and manages the virtual volume 21 used by the I / O server 18. The volume management server 15 includes a control unit 15a, a management LAN IF 15b, and a storage unit 15c. The control unit 15a controls the operation of the volume management server 15, and is, for example, a CPU (central processing unit). The management LAN IF 15 b is a communication interface for connecting to the management LAN 27. The storage unit 15c stores a program for operating the control unit 15a, a program according to the present embodiment, a table to be described later, and the like. As the storage unit 15c, for example, a ROM (read only memory), a RAM (random access memory), a hard disk, a flash memory, or the like can be used.

  The I / O server 18 controls input from the client device 11 and output to the client device 11. The I / O server 18 includes an external communication IF 18a, a control unit 18b, a storage unit 28c, a cache memory (hereinafter referred to as “cache”) 18d, a storage communication IF 18e, and a management LAN IF 18f. The external communication IF 18 a is a communication interface for connecting to the LAN 13. The control unit 18b controls the operation of the I / O server 18 and controls a virtual volume obtained by virtualizing a storage disk, and is, for example, a CPU (central processing unit). The storage unit 18c stores a program for operating the control unit 18b, a program according to the present embodiment, a table to be described later, a virtual volume, and the like. As the storage unit 18b, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), a hard disk, a flash memory, or the like can be used. The storage communication IF 18 e is a communication interface for connecting to the I / O LAN 28. The management LAN IF 18 f is a communication interface for connecting to the management LAN 27.

  The storage device 25 includes a plurality of disks 26. The storage device 25 is connected to a management LAN 27 and an I / O LAN 28. The management LAN 27 is a network used for monitoring commands and loads among the volume management server 15, the I / O server 18, and the storage device 25. The I / O LAN 28 is a network for transferring data stored in the storage device 25 or data stored in the storage device 25. The management LAN 27 and the I / O LAN 28 are not limited to LANs, and may be networks such as a SAN (Storage Area Network), for example.

  A program that realizes processing described in the following embodiment may be stored in the storage units 15c and 18b, for example, via the LAN 13 from the program provider side. Moreover, the program which implement | achieves the process demonstrated by the following embodiment may be stored in the portable storage medium marketed and distribute | circulated. In this case, the portable storage medium may be set in a reading device (not shown) of the storage system 14, and the program may be read and executed by the control units 15a and 18b. As the portable storage medium, various types of storage media such as a CD-ROM, a flexible disk, an optical disk, a magneto-optical disk, an IC card, and a USB memory device can be used. A program stored in such a storage medium is read by a reading device.

  FIG. 3 shows a block diagram of the volume management server 15 and the I / O server 18. The volume management server 15 includes a volume management unit 16 and a load monitoring unit 17. The volume management unit 16 manages the virtual volume 21 of each I / O server 18. The load monitoring unit 17 monitors the load information of each storage device 25 and the load of the volume in the storage device 25.

  The volume management unit 16 includes a replica table 31, a write hold table 32, an I / O periodic table 33, a read / write access load correspondence table 34, and a read / write disk access load correspondence table 35. The replica table 31, the write hold table 32, the I / O periodic table 33, the read / write access load correspondence table 34, and the read / write disk access load correspondence table 35 are stored in the storage unit 15c of the volume management server 15.

  The I / O server 18 includes an I / O management unit 20 and a virtual volume 21. The I / O management unit 20 controls access from the client device 11 to the virtual volume 21 and output from the virtual volume 21 to the client device 11. The virtual volume 21 is stored in the storage unit 18 c provided in the I / O server 18.

  The I / O management unit 20 includes a replica completeness table 36, a disk I / O periodic table 37, and a read / write position access load correspondence table 38. The replica completeness table 36, the disk I / O periodic table 37, and the read / write position access load correspondence table 38 are stored in the storage unit 18c of the I / O server 18.

  FIG. 4 shows an example of the replica table 31. The replica table 31 is a table indicating in which storage device 25 the replica of the virtual volume 21 is created or in what state the replica is.

  The replica table 31 includes data items of “virtual volume ID”, “storage ID 1”, “storage ID 2”,. In the “virtual volume ID”, identification information for identifying the virtual volume 21 is stored. In “Storage ID 1”, “Storage ID 2”,..., The degree of completion of the replica in the storage device 25 corresponding to the virtual volume 21 is set. Here, if the degree of completion of the replica is 100%, “Created” is set. When the degree of completion of the replica is “0 <completeness <100”, “Copying” is set. When the degree of completion of the replica is 0%, “None” is set. If the replica is being deleted, “Deleting” is set.

  FIG. 5 shows an example of the write hold table 32. The write hold table 32 is a table for managing write target data held in the cache 18d. The write hold table 32 includes data items of “virtual disk ID”, “storage ID”, “location”, and “date”. In the “virtual disk ID”, identification information for identifying the virtual volume 21 is stored. In “Storage ID”, information (storage ID) for identifying the storage device 25 corresponding to the virtual volume 21 is stored. “Position” stores a write position in the storage 25 where data is written. In “Date”, the written date is stored.

  FIG. 6 shows an example of the I / O periodic table 33. The I / O periodic table 33 is a table in which the number of accesses (read / write) of each disk of each storage device 25 per unit time is tabulated. The I / O periodic table 33 is created for each disk of the storage device 25.

  The I / O periodic table 33 includes data items of “time”, “read”, and “write”. In “Time”, time at a predetermined interval is stored. “Read” stores the number of read accesses at that time. “Write” stores the number of write accesses at that time.

  FIG. 7 shows an example of the read / write access load correspondence table 34. The read / write access load correspondence table 34 is a table in which the disk ID of the read / write disk access load correspondence table 35 is associated with the disk ID of the read / write position access load correspondence table 38. The read / write disk access load correspondence table 35 and the read / write position access load correspondence table 38 will be described later.

  FIG. 8 shows an example of the read / write disk access load correspondence table 35. The read / write disk access load correspondence table 35 is a table for counting the number of times of read access and write access to each disk existing in all the storages 25.

  The read / write disk access load correspondence table 35 is a storage device unit, and includes data items of “disk ID”, “Read access frequency”, and “Write access frequency”. In the “disk ID”, identification information for identifying the disk 26 in the storage device 25 is set. In “Read access frequency”, the read access frequency for the disk is set. In “Write access frequency”, the write access frequency for the disk is set.

  FIG. 9 shows an example of the replica completeness table 36. The replica completeness table 36 is a table showing the completeness of the replica for each storage device 25 of the virtual volume for each position. The replica completeness table 36 stores “location” where the replica is stored in the storage device 25, “replica completeness” of each storage device, and “state” of the replica stored in the storage device 25.

  The replica completeness is a value indicating how much the replica is completed. The “state” of the replica has four states of “None”, “Copying”, “Created”, and “Deleting”. “None” indicates that a replica has not been created. “Copying” indicates that it is before issuing a write process to the replica creation destination. “Created” indicates that replica creation has been completed. “Deleting” indicates that the replica is being deleted.

  FIG. 10 shows an example of the disk I / O periodic table 37. It is the table | surface which totaled the access count of each disk of the storage 25 per unit time. The disk I / O periodic table 37 is a table in which the number of times the I / O server has accessed each disk 26 of the storage device 25 at a certain time is tabulated. The disk I / O periodic table 33 is created for each disk of the storage device 25.

  The disk I / O periodic table 33 includes data items of “time”, “read”, and “write”. In “Time”, time at a predetermined interval is stored. “Read” stores the number of read accesses at that time. “Write” stores the number of write accesses at that time.

  FIG. 11 shows an example of the read / write position access load correspondence table 38. The read / write position access load correspondence table 38 is a table that manages the read / write access load for each position of the disk 26 to which the I / O server 18 is accessing. Here, the read / write position on the disk 26 indicates a position relative to the head position of the disk 26 for reading / writing data.

  The read / write position access load correspondence table 38 is provided for each disk, and includes data items of “disk ID”, “Read access frequency”, and “Write access frequency”. In the “disk ID”, identification information for identifying the disk 26 in the storage device 25 is set. In “Read access frequency”, the Read access frequency for the disk 26 is set. In the “Write access frequency”, the write access frequency for the disk 26 is set.

  FIG. 12 shows an example of the storage capacity table 39. The storage capacity table 39 is a table in which the capacity of each storage device 25 is stored. The storage capacity table 39 is a list of capacities used by the storage apparatus acquired from each storage apparatus 25 by the load monitoring unit 17 of the volume management server 15.

  FIG. 13 shows an example of the load statistical information 40. The load statistical information 40 is a table storing read / write (I / O per second) and bandwidth usage rate per unit time for each storage device 25. The volume management server 15 acquires the load statistical information 40 from each storage device 25 using the load monitoring unit 17.

  In this embodiment, the storage system 14 to which the storage device is added has two or more storage devices. For example, the storage system 14 includes two storage apparatuses A and B. It is assumed that the disk in the storage apparatus A maintains a mirror state (that is, two replicas) with the disk in the storage apparatus B. A storage device added to the storage system 14 is a storage device C.

  (1) Add the storage device C to the storage system 14.

(2) The following processing is performed when the user reads / writes data using the client 11.
(2-1) Using the load monitoring unit 17, select a storage device 25 with a low load.
(2-2) Data is read from the selected storage device 25.
(2-3) Return the read data to the client 11.

  (3) The data held in the cache at the time of data reading is written to the disk of the storage device C at a time when the network bandwidth of the storage device C is not squeezed (at this time, the replica becomes 3). As described above, the data is moved to the storage device C with a small load, so that the reading performance in the next reading can be improved.

  (4) If only read is performed at a high frequency and the read performance of any replica is degraded, a replica is created if there is a newly added storage device.

  (5) When the free capacity of the storage device 25 is small (a certain threshold is set, for example, 80%), the free capacity is increased by deleting those having 3 or more replicas. Here, using the I / O periodic table 33, replicas of volumes that have a large number of read periods after writing are excluded from deletion targets. When the number of replicas is 3 or more and write processing is frequently performed, the replica of the storage device with a high load is deleted on condition that at least two replicas are maintained.

Below, the detail of this embodiment is demonstrated.
A storage device to be added to the storage system 14 is a storage device C (25c). In the present embodiment, explanation will be given in the case of a new addition of the storage apparatus 25, but rebalancing can be performed with the same algorithm even after the operation is started.

(1) Addition of Storage Device C (25c) The storage device C (25c) is installed in the storage system 14 and connected to the I / O LAN 27 so that all clients 11 can access it.

(2) Data Reading FIG. 14 shows a data reading flow in this embodiment. Hereinafter, the operations of the volume management server 15 and the I / O server 18 will be described with reference to FIG.

[Select Read Storage]
The load monitoring unit 17 of the volume management server 15 monitors the load status of each storage device 25 and the load of the volume in the storage device (S11). Here, the load monitoring unit 17 acquires the load statistical information 40 transmitted from the storage device 25.

  The volume management server 15 periodically transmits the load statistical information acquired from the storage device 25 to the I / O server 18 (S12).

[Retention of read data cache]
When there is a data read request from the client 11, the I / O server 18 reads the data from the disk 26 in the storage device 25 (S21). When it is determined that a plurality of replicas exist with reference to the replica completeness table 36, the I / O server 18 selects the storage device 25 having the lowest access load or the storage device 25 having the disk 26 having the lowest access load. (S22). For example, the I / O server 18 uses the bandwidth usage rate or I / Ops of the load statistical information received from the volume management server 15 to select the storage device 25 with the lowest access load. Alternatively, the I / O server 18 uses the read / write position access load correspondence table 38 to select the storage device 25 having the disk 26 with the lowest access load.

  The I / O server 18 reads data from the storage device 25 selected in S21 or the storage device 25 having the disk 26 having the lowest access load (S23). At this time, the I / O server 18 updates the read access frequency at the position where the disk 26 is read in the read / write position access load correspondence table 38 (S24).

  When reading data, the I / O server 18 holds the read data in the cache 18d in order to improve the performance when the same data is read again (S25). The I / O server 18 transmits the data held in the cache 18d to the client 11 (S26). Thereby, reading of data is completed.

  When reading a file on the file system of the storage device 25, the I / O server 18 pre-reads data and holds the pre-read data in the cache 18d.

[Use read data cache]
The I / O server 18 uses the data held in the cache 18d at the time of reading (including prefetched data if it exists) in the background of the data reading request from the client 11, and writes a request to the replica of the added storage device 25c. (S27). By using the data held in the cache 18d, reading for copying is not performed, so that the network load required for data copying for data rebalancing can be reduced to ½.

  If data is held in the cache 18d before the replica start instruction from the volume management server 15, the I / O server 18 performs the following processing. The I / O server 18 writes the data to the replica creation destination when a replica creation destination to be newly added to the virtual volume 21 is selected, as will be described from S56 onward in FIG. Thus, efficient replica creation can be performed by using the data held in the cache 18d.

(3) Data Write FIG. 15 shows a data write flow in this embodiment. Hereinafter, the operations of the volume management server 15 and the I / O server 18 will be described with reference to FIG.

  When there is a data write request from the client 11, the I / O server 18 starts a data write process to the real disk 26 of the virtual volume 21 (S31). At this time, in order to maintain data consistency, the I / O server 18 reads the replica completeness table 36 (S33), and acquires information on the storage device (target storage device) where the write target replica exists. (S32).

  The I / O server 18 confirms the load status of the target storage device 25 using the load statistical information received from the volume management server 15 in S12 (S34). The I / O server 18 uses the load statistical information to determine whether there is no load on all the target storage devices 25, that is, whether the bandwidth usage rate of all the target storage devices 25 is lower than the threshold (S35). When the bandwidth usage of all the target storage devices 25 is lower than the threshold (“Yes” in S35), the I / O server 18 writes data to all the target storage devices 25 (S42). The I / O server 18 updates the read / write position access load correspondence table 34 for the target storage apparatus 25 that has written data (S43). The I / O server 18 notifies the write request source client 11 that the data writing has been completed (S44).

[Write to multiple replicas]
If the load on any of the target storage devices 25 is higher than the threshold (“No” in S35), the I / O server 18 performs the following. That is, the I / O server 18 uses the load statistical information or the read / write position access load correspondence table 38 to write the write target data to the storage device 25 with the lowest load or the disk with the lowest load. At the time when one writing is completed, the I / O server 18 notifies the client 11 as the write request source that the data writing has been completed (S36).

[Countermeasures for write pending storage]
The I / O server 18 needs to write to the storage device 25 elsewhere, but writing data at a high load causes degradation in performance. For this reason, the I / O server 18 holds the write data to the storage device 25 in which the height of the load exceeds the threshold and there is no I / O margin in the cache 18d, and temporarily holds the data write. The data held in the cache 18d is not discarded. This will be described below.

  When the data to be written to the storage device 25 is suspended, the I / O server 18 notifies the volume management server 15 of the suspension information (S37). The hold information includes the storage ID, disk ID, disk write position, and write data of the target storage device 25 that has not been written. The volume management server 15 updates the write hold table 32 using the hold information notified from the I / O server 18 (S38). The I / O server 18 holds the write data in the cache 18d (S39).

  At the time of reading data from the other I / O server 18, the other I / O server 18 refers to the write hold table 32 and performs a process of reading the data into the storage device 25 where the latest data and the written data exist. Do. Thereby, the latest data can be read.

  Thereafter, the I / O server 18 uses the load statistics information or the read / write position access load correspondence table 38 to hold the I / O server 18 in the cache 18d when the load on the storage device 25 falls below the threshold. The written data is written into the storage device 25 (S40). At this time, the I / O server 18 updates the write access frequency at the position written in the disk 26 of the storage device 25 in the read / write position access load correspondence table 38 (S41). When the writing to the storage device 25 is completed, the I / O server 18 issues a completion notification to the volume management server 15. Thereby, the same data can be read from all the storage devices 25.

  In addition, when the I / O server 18 goes down while the write target data is held in the cache 18d, the write target data held in the cache 18d is lost, so it is possible to manage which replica is the latest. Can not. Therefore, when the I / O server 18 is started after being down, the volume management server 15 checks whether there is data to be written in the cache 18d in the write hold table 32, and creates a replica if necessary.

  The storage devices A and B and the added storage device C hold the same data by completing the writing of the data held in the cache 18d to the storage device 25 to the storage device 25 performing the replica creation. . At this time, the storage apparatus A and the storage apparatus B are maintained in a mirror state. In addition, for the storage device C, only the data that has been written is written.

  Here, in the added storage device C, the data held in the cache 18d by read / write processing (data that is frequently accessed) can be kept in the mirror state, but the entire disk can be mirrored. No. Therefore, at this time, the entire original disk of the replica cannot be deleted. However, in this case, since the read data is in the mirror state (because a replica is made), the reading of the data of the I / O server 18 to the storage device 25 can be distributed. As a result, the access (load) to the storage device that is the read destination is not concentrated on one, so that the performance of the storage system 14 can be improved.

(4) Access load distribution method by replica creation [Replica creation method]
16 and 17 show a replica creation processing flow in the present embodiment. Hereinafter, the operations of the volume management server 15 and the I / O server 18 will be described with reference to FIGS. 16 and 17.

  The volume management server 15 reflects the information obtained from the load monitoring unit 17 in the read / write disk access load correspondence table 35 as needed. Further, the volume management server 15 uses the read / write disk access load correspondence table 35 and the read / write position access load correspondence table 38 periodically sent from all the I / O servers 18 to read / write access load. A correspondence table 34 is created.

  First, the volume management server 15 uses the read / write access load correspondence table 34 to determine whether or not the access load (read / write access frequency) exceeds a preset threshold value for each storage device (S51). ).

  When the access load exceeds the threshold value (“Yes” in S51), the volume management server 15 refers to the read / write access frequency of the disk in the read / write access load correspondence table 34, and the position of the data with a high access load. Is specified (S52). That is, the volume management server 15 uses the read / write access load correspondence table 34 for the data at which position on which disk in the storage device 25 in which the access load (read / write access frequency) exceeds the threshold. Determine if there are many read / writes.

Furthermore, the volume management server 15 determines a replica creation destination using the read / write access load correspondence table 34 (S53). Here, the volume management server 15 uses the read / write access load correspondence table 34 to select the storage device 25 with the lowest access load. Furthermore, the volume management server 15 selects a storage device with the largest free capacity from the selected storage devices 25 as a replica creation destination.
Then, the volume management server 15 notifies the I / O server 18 of the replica creation destination (S54).

  Upon receiving notification of the replica creation destination from the volume management server 15, the I / O server 18 creates a replica completeness table 36 for the virtual volume 21 (S55). Note that the Read process (FIG. 14) and the write process (FIG. 15) occur between S55 and S56.

  After creating the replica completeness table 36, the I / O server 18 adds a replica creation destination disk to one of the real disks of the virtual volume 21 (S56). The I / O server 18 notifies the volume management server 15 to start writing the data held in the cache to the replica creation destination disk (S57).

  When the volume management server 15 receives a notification from the I / O server 18 to start writing the data held in the cache to the replica creation destination disk, it updates the corresponding part of the replica table 31 to “Copying” (S58). ).

[Write data to replica]
If the replica source data exists in the cache 18d ("Yes" in S61), the I / O server 18 uses the cache 18d following the data read processing from the client 11 to the replica storage device 25. Then, the replica creation process is started (S62). Further, regarding the data held in the cache 18d before the replica creation request is received from the volume management server 15, the I / O server 18 updates the replica completeness table 36 and stores the data in the replica creation target disk. Write. By using the data already held in the cache 18d, the replica update process can proceed.

  The I / O server 18 uses the information (replica creation destination and write time) notified from the volume management server 15 to write the data held in the cache 18d at the time of data read to the replica destination at a preset time. (S63). After completion of the writing, the I / O server 18 updates the created disk 26 of the storage device 25 to “Created” in the replica completeness table 36 (S64).

  In the replica completeness table 36, when the replica completeness reaches 100%, the I / O server 18 notifies the volume management server 15 that the replica creation is completed (S65). The volume management server 15 updates the corresponding part of the replica table 31 to “Created” (S66).

  The I / O server 18 notifies the volume management server 15 to start processing to write the data held in the cache 18d to the replica creation destination disk (S67).

[Replica deletion method]
When there are frequent write requests from the client 11, writing to all replicas occurs frequently, resulting in a high network bandwidth load. When writing frequently occurs, the I / O server 18 determines whether or not to delete the replica by the following processing.

(I) If the I / O server 18 uses the disk I / O periodic table 37 for the data to be written and determines that there is a large amount of reading period after that from the amount of data read / written, Do not delete the replica.
(Ii) If the I / O server 18 uses the disk I / O periodic table 37 and determines that the write period continues for a certain period, the I / O server 18 deletes the replica.

  18 and 19 show a replica deletion processing flow in the present embodiment. Hereinafter, the operations of the volume management server 15 and the I / O server 18 will be described with reference to FIGS. 18 and 19.

  When it is determined that any of the following conditions (S71, S72) is met, the volume management server 15 notifies the I / O server 18 of the start of replica deletion processing (S73). The condition of S71 is that the volume management server 15 determines from the write access frequency in the read / write access load correspondence table 34 that the write access frequency to the data in the replica of any virtual volume has exceeded the threshold. . Further, the condition of S72 is that the volume management server 15 uses the storage capacity table 39 to determine that the free capacity of the storage device 25 is below the threshold when creating a replica of any virtual volume.

-Determination method of replica to be deleted-
In the replica deletion processing, the volume management server 15 refers to the replica table 31 and determines whether there are three or more replicas of any virtual volume that is “Created” (S74). When the number of replicas is less than 3 (“No” in S74), the volume management server 15 does not execute the replica deletion process.

  When the number of replicas is 3 or more, the volume management server 15 uses the write access frequency item in the read / write access load correspondence table 34 to select a replica in the storage apparatus 25 with the highest access load. (S75).

  The volume management server 15 associates the read / write access load correspondence table 34 with the I / O periodic table 33 and exceeds the threshold for the disk where the selected replica exists within a predetermined time from the current time. It is determined whether there is a read access (S76).

  When it is determined that read access exceeding the threshold is not made within a predetermined time from the current time for the selected replica (“Yes” in S76), the volume management server 15 deletes the replica for the virtual volume 21 Do not execute. At this time, the volume management server 15 searches for a replica of another virtual volume 21 (S78).

If there is a replica of another virtual volume (“Yes” in S79), the volume management server 15 performs the processing subsequent to the processing in S73 for the other virtual volume.
When there is no replica of another virtual volume (“No” in S79), the volume management server 15 notifies the I / O server 18 to dynamically lower the replica completeness threshold used in S84 by a predetermined value. (S80). Then, the volume management server 15 performs the processing subsequent to S73 again for the virtual volume. Upon receiving the notification, the I / O server 18 dynamically lowers the replica perfection threshold used in S84 by a predetermined value.

-Execution of replica deletion When it is determined that the read access exceeding the threshold is made within a predetermined time from the current time ("No" in S76), the volume management server 15 determines that the replica table 31 Update the relevant part of to "Deleting". Thereafter, the volume management server 15 notifies the deletion information (virtual volume ID, data arrangement location) about the selected replica to the I / O server 18 that is accessing the relevant replica (S77).

  The I / O server 18 receives deletion information about the selected replica from the volume management server 15. The I / O server 18 uses the replica completeness table 36 to determine whether there are three or more replicas with 100% replica completeness related to the notified virtual volume ID (S81).

  In the replica completeness table 36, when three or more replicas having a replica completeness of 100% do not exist (“No” in S81), the I / O server 18 applies the replicas to the replicas being created. Continue creation. After completing the replica, the I / O server 18 deletes the replica of the storage device 25 with a high load selected in S75 (S83), and ends the replica deletion processing.

  In the replica completeness table 36, the I / O server 18 determines whether or not the total replica completeness of the notified virtual volume ID is equal to or less than a predetermined threshold (S84). If the completeness of all replicas of the notified virtual volume ID is less than or equal to a predetermined threshold (for example, 80%) (“Yes” in S84), the I / O server 18 sends the volume management server 15 to another virtual volume. A request for changing a replica deletion target is made (S85).

  If there is a replica of another virtual volume (“Yes” in S79), the volume management server 15 performs the processing subsequent to the processing in S73 for the other virtual volume. When there is no replica of another virtual volume (“No” in S79), the volume management server 15 notifies the I / O server 18 to dynamically lower the replica completeness threshold used in S84 by a predetermined value. (S80). Then, the volume management server 15 performs the processing subsequent to S73 again for the virtual volume. Upon receiving the notification, the I / O server 18 dynamically lowers the replica perfection threshold used in S84 by a predetermined value. As a result, the volume management server 15 can request the I / O server 18 to delete the replica again.

  If the completeness of all replicas of the notified virtual volume ID is higher than a predetermined threshold (for example, 80%) (“No” in S84), the I / O server 18 performs the following processing. That is, the I / O server 18 uses the replica creation destination information notified from the volume management server 15 in S54 to write the data held in the cache 18d at the time of data reading to the replica creation destination, and the replica completeness is set to 100%. (S86).

  In the replica completeness table 36, when there are three or more replicas having a replica completeness of 100% related to the notified virtual volume ID (“Yes” in S81), or after the processing of S86 is completed, the I / O server 18 Performs the following processing. That is, the I / O server 18 determines whether the free capacity of the storage device 25 has fallen below the threshold (S87). When the free space of the storage exceeds the threshold (“No” in S87), the replica deletion process is terminated.

  When the free capacity of the storage device 25 is lower (“Yes” in S87), the I / O server 18 uses the read / write position access load correspondence table 38 to read the read access frequency from the replica with 100% completeness of replica. A replica whose value is below the threshold is selected (S88). At this time, the I / O server 18 selects the replica of the storage device 25 with the highest load among the storage devices 25 holding the replica having a replica completeness of 100%. The I / O server 18 deletes the selected replica (S89).

  The I / O server 18 notifies the volume management server 15 of the completion of the replica deletion process (S90). When receiving the notification of completion of the replica deletion process from the I / O server 18, the volume management server 15 updates the corresponding part of the replica table 31 from “Deleting” to “None” (S91), and ends the replica deletion process.

[Predictive replica creation process using I / O periodic table]
When there are a large number of disks with low access loads that cannot be captured by “(4) Access load distribution method by replica creation”, the load may increase when viewed in units of storage devices. Therefore, in the data rebalancing in such a situation, a method of using the I / O periodic table and creating a replica to distribute the load is used.

  FIG. 20 shows a predicted replica creation processing flow using the I / O periodic table. The I / O server 18 aggregates the input / output (I / O) (that is, the number of read accesses and the number of write accesses) at regular intervals (cycles) in a predetermined period for each disk in the storage device 25, Store in the disk I / O periodic table 37. In the case where the fixed interval (cycle) is a business in units of one week, the I / O server 18 can also collect data in one week. The process of FIG. 20 may be executed in the process of S54 of FIG.

  The volume management server 15 collects the disk I / O periodic table 37 from all the I / O servers 18 at a predetermined time. The volume management server 15 uses the collected all-disk I / O periodic table 37 to create the I / O periodic table 33 by counting the number of read accesses and the number of write accesses for each disk time of the storage device. .

  The volume management server 15 aggregates the I / O per time (total number of read accesses and write accesses) for each storage device 25 from the created I / O cycle table 33 at a constant cycle. The access load is calculated (S101). For example, the volume management server 15 calculates the access load of the storage device 25 every hour for one day. Here, the access load refers to the total number of read accesses and write accesses.

  Next, the volume management server 15 calculates an average value of access loads for all the storage devices 25 (S102). Thereafter, the volume management server 15 determines whether or not the difference between the average value and the access load of each storage device 25 exceeds a preset threshold during the tabulated time period (S103). If the difference in the success load does not exceed the threshold (“No” in S103), the volume management server 15 ends the prediction process.

  When the difference in the success load exceeds the threshold (“Yes” in S103), the volume management server 15 performs the following process. That is, the volume management server 15 determines the difference between the access load of each disk 26 and the average value of the access loads of all the disks 26 belonging to the storage device 25 in smaller time units with respect to the corresponding time of the corresponding storage device 25. Ask for. The finer time unit is, for example, 1 minute unit.

  The volume management server 15 determines whether the difference is below a preset threshold value (S104). If the difference is below the preset threshold (“Yes” in S104), the volume management server 15 uses the I / O periodic table 33 to read the most read disk 26 in the storage device 25. A disk 16 with a lot of processing is selected. The volume management server 15 creates a replica on the selected disk 26 before the corresponding time and in a time zone where the access load of the storage device 25 does not exceed a preset threshold (S105).

  In the following, an example of predicted replica creation processing for data having a high access load in the storage device 25 unit and a low access load in the disk unit using the I / O periodic table 33 will be described. First, the processing of S101 to S103 in FIG. 20 will be described with reference to FIGS.

  FIG. 21 shows a work table (A) in which access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the storage device A are totaled every hour, and a graph (B) thereof. Show. FIG. 22 shows a work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the storage apparatus B are counted every hour, and a graph (B) thereof. Show. FIG. 23 shows a work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the storage device C are counted every hour, and a graph (B) thereof. Show. FIG. 24 shows an average value of the access frequency for each hour of the work table (A) shown in FIGS. 21 to 23 and a graph (B) thereof.

  The volume management server 15 collects the disk I / O periodic table 37 from all the I / O servers 18 and creates the I / O periodic table 33, and then, as shown in FIGS. , B, and C are calculated. After calculating the access load, the volume management server 15 calculates the average access load for the three storage apparatuses A, B, and C as shown in FIG. Here, a description will be given focusing on the access load in the time zone 4:00 to 5:00.

  As shown in FIG. 21, the access load of the storage apparatus A is 545. As shown in FIG. 22, the access load of the storage apparatus B is 5435. As shown in FIG. 23, the access load of the storage apparatus C is 23. As shown in FIG. 24, the average value of the access load is 2001.

  The volume management server 15 calculates the difference between the access load of each storage device 25 and the average value. Here, the difference between the storage device B and the average value is 5435−2001 = 3434. The threshold value set in advance is set to 2000. Since the difference (3434) between the access load and the average value of the storage device B exceeds the threshold value (2000), the volume management server 15 executes the processing of S104-S105 in FIG. This will be described with reference to FIGS.

  FIG. 25 shows a work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the disk B1 of the storage device B are counted every minute, and its graph ( B). FIG. 26 shows a work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the disk B2 of the storage device B are counted every minute and its graph ( B). FIG. 27 shows a work table (A) in which the access frequencies obtained by adding the read frequency and the write frequency of the I / O periodic table 33 for the disk B3 of the storage apparatus B are counted every minute, and its graph ( B). FIG. 28 shows an average value of access frequencies per minute of the work table (A) shown in FIGS. 25 to 27 and a graph (B) thereof.

  The volume management server 15 performs the following processing for the corresponding time (time zone 4:00 to 5:00) of the storage device 25 where the difference between the access load of the storage device 25 and the average value exceeds the threshold value. That is, the volume management server 15 obtains the difference between the access load of each disk belonging to the storage device 25 and the average value of the access loads of all the disks in units of one minute. Here, it is assumed that there are three disks B1, B2, and B3 in the storage apparatus B, and explanation will be given focusing on the time zone 4:31 to 4:32.

  First, the average value of the access load of the storage device B per minute is 5435 / 60≈91. Furthermore, since there are three disks in the storage apparatus B, the average value of the access load for each minute of each disk is 91 / 3≈31. Here, the access loads of the disks B1, B2, and B3 in the time zone of 4:31 to 4:32 are 27, 24, and 29, respectively. The difference between the access load and the average value of the disks B1, B2, and B3 is 4, 7, and 2, respectively. If the preset threshold value is 10, the difference between the access load and the average value of the disks B1, B2, and B3 are all below the threshold value. In all other time zones, the difference between the access load and the average value of the disks B1, B2, and B3 is all below the threshold value. From this, it can be seen that the access load on the storage device B is high, but the access load on each disk B1, B2, B3 in the storage device is low. In this example, in the replica creation process, the storage apparatus B is excluded from the rebalance target, and the rebalance for distributing the access load of the storage apparatus B cannot be performed.

  Therefore, the volume management server 15 selects the disk with the most read processing among the disks B1, B2, and B3 by using the predicted replica creation process using the I / O periodic table 33. The volume management server 15 instructs the I / O server 18 to create a replica of the data on the selected disk 26.

  By performing rebalancing between storage apparatuses as described above, data is not lost even if an abnormality occurs in the storage apparatus or a disk in the storage apparatus. For example, if the storage device B fails when the data is transferred from the storage device A to the storage device C, the redundancy of the replica decreases, but the data can be obtained by combining the storage devices A and C. Can guarantee up-to-date. Further, when the volume management server 15 recognizes that the replica redundancy has decreased, the volume management server 15 forcibly performs the replica creation processing so that the replica redundancy is 2.

  In an environment such as a large-scale cloud having a plurality of storage devices 25 and I / O servers 18, the bandwidth between the I / O servers 18 and the storage devices 25 is important. By effectively using the data necessary for reading / writing corresponding to the input / output request held in the cache, the storage device 25 can be rebalanced without imposing a load on the network bandwidth.

  According to the present embodiment, when a storage device is added, the rebalancing process between the storages can be efficiently performed without imposing a load on the network bandwidth by effectively using the access request from the client device. it can. In addition, the input / output performance of the storage system can be fully utilized. Therefore, if the storage system in this embodiment is used, the efficiency of business and services can be improved without affecting business.

  In addition, even when an abnormality occurs in a storage device or a disk included in the storage system in this embodiment, data is always redundant, so that business can be continued without losing data.

  The present invention is not limited to the above-described embodiment, and various configurations or embodiments can be taken without departing from the gist of the present invention.

DESCRIPTION OF SYMBOLS 1 Storage system 2 Storage apparatus 3 Cache memory 4 Access control part 5 Writing part 6 Load monitoring part 7 Replica deletion part 8 Replica creation part 9 Information processing terminal 11 Client apparatus 12 Communication interface 13 LAN
14 Storage System 15 Volume Management Server 16 Volume Management Unit 17 Load Monitoring Unit 18 I / O Server 18d Cache Memory 20 I / O Management Unit 21 Virtual Volume 25 Storage Device 26 Disk 27 Management LAN
28 LAN for I / O
31 replica table 32 write pending table 33 I / O periodic table 34 read / write access load correspondence table 35 read / write disk access load correspondence table 36 replica completeness table 37 disk I / O periodic table 38 read / write position access load correspondence table

Claims (18)

A plurality of storage devices for storing data;
Cache memory to hold data,
When there is an access request for reading target data or writing target data from an information processing terminal, an access control unit that accesses any of the storage devices and stores the target data in the cache memory;
A writing unit for writing the target data stored in the cache memory to a storage device in which the target data is not stored among the plurality of storage devices;
A storage system comprising: The storage system further includes:
A load monitoring unit for monitoring an access load by reading or writing to the storage device;
The access control unit, when receiving the access request from the information processing terminal, accesses the storage device with the least access load among the storage devices based on the monitoring result, and caches the target data. The storage system according to claim 1, wherein the storage system is stored in a memory. When there is a read request to the target data from the information processing terminal, the access control unit is a storage device with the least access load among the storage devices that store the target data based on the monitoring result The storage system according to claim 2, wherein the target data is acquired from the storage, the target data is stored in the cache memory, and the target data is transmitted to the information processing terminal. The access control unit writes the target data to the storage device with the least access load among the storage devices based on the monitoring result when there is a request to write the target data from the information processing terminal, Store the target data in the cache memory,
The storage system according to claim 2, wherein the writing unit writes the target data stored in the cache memory to a storage device other than the storage device that performed the writing. The storage system further includes:
When the write access load of a replica of a virtual volume obtained by virtualizing the volumes of the plurality of storage devices exceeds a threshold value, or when the free capacity of the storage device is less than a first threshold value in the creation of the replica, the virtual 2. The storage system according to claim 1, further comprising: a replica deletion unit that deletes the replica stored in one of the storage devices when there are three or more replicas of the volume in the plurality of storage devices. The replica deletion unit selects the replica of the storage device with the least access load from the storage devices including the replica based on the monitoring result, and indicates the access frequency indicating the access frequency to the storage device that is aggregated at each time The selected replica is deleted when it is determined that there is no read access with an access frequency exceeding a second threshold after the current time on the selected replica based on frequency information. 6. The storage system according to 5. The storage system further includes:
The difference between the access load in any time zone among the access loads for each time for any of the storage devices and the average of the access loads in the time zone for all the storage devices exceeds the third threshold The time zone is subdivided, the access load on any disk of the storage device in any one of the subdivided time zones, and the total disk of the storage device If the difference from the average access load of any of the subdivided time zones is smaller than the fourth threshold, the access load of any of the storage devices is the fifth before the subdivided time zone The storage system according to claim 1, further comprising: a replica creation unit that creates a replica on a disk having the highest access load during a time that does not exceed the threshold. On the computer,
When there is an access request for reading target data or writing target data from an information processing terminal, it accesses any one of a plurality of storage devices for storing the data, and stores the target data in a cache memory Stored in
A data rebalancing program for executing a process of writing the target data stored in the cache memory to a storage device in which the target data is not stored among the plurality of storage devices. In addition to the computer,
A process of monitoring an access load by reading or writing on the storage device;
When there is an access request from the information processing terminal, based on the monitoring result, the storage device with the least access load is accessed among the storage devices, and the target data is stored in a cache memory. 9. The data rebalancing program according to claim 8, wherein When there is a read request to the target data from the information processing terminal, the target data is acquired from the storage device with the least access load among the storage devices that store the target data based on the monitoring result The data rebalancing program according to claim 9, wherein the target data is stored in the cache memory, and the target data is transmitted to the information processing terminal. When there is a request for writing target data from the information processing terminal, based on the monitoring result, the target data is written to the storage device with the least access load among the storage devices, and the target data is stored in the cache memory. Stored in
The data rebalancing program according to claim 9, wherein the target data stored in the cache memory is written to a storage device other than the storage device that performed the writing. In addition to the computer,
When the write access load of a replica of a virtual volume obtained by virtualizing the volumes of the plurality of storage devices exceeds a threshold value, or when the free capacity of the storage device is less than a first threshold value in the creation of the replica, the virtual The data rebalancing program according to claim 8, wherein if there are three or more replicas of a volume in the plurality of storage devices, a process of deleting the replica stored in any of the storage devices is executed. Based on the monitoring results, select the storage device replica with the least access load among the storage devices including the replica, and based on the access frequency information indicating the access frequency to the storage device counted for each time, The data replica according to claim 12, wherein if the selected replica is determined not to have read access with an access frequency exceeding a second threshold after the current time, the selected replica is deleted. Balancing program. In addition to the computer,
The difference between the access load in any time zone among the access loads for each time for any of the storage devices and the average of the access loads in the time zone for all the storage devices exceeds the third threshold The time zone is subdivided, the access load on any disk of the storage device in any one of the subdivided time zones, and the total disk of the storage device If the difference from the average access load of any of the subdivided time zones is smaller than the fourth threshold, the access load of any of the storage devices is the fifth before the subdivided time zone The data rebalancing program according to claim 8, wherein a process of creating a replica on a disk having the largest access load is executed during a time when the threshold is not exceeded. A method of rebalancing data between storage devices executed by a computer,
The computer
When there is an access request for reading target data or writing target data from an information processing terminal, it accesses any one of a plurality of storage devices for storing the data, and stores the target data in a cache memory Stored in
A data rebalancing method for executing processing for writing the target data stored in the cache memory to a storage device in which the target data is not stored among the plurality of storage devices. The computer further includes:
Monitor the access load by reading or writing to the storage device,
When there is an access request from the information processing terminal, based on the monitoring result, the storage device with the least access load is accessed among the storage devices, and the target data is stored in a cache memory. The data rebalancing method according to claim 15, characterized in that: The computer
When there is a read request to the target data from the information processing terminal, the target data is acquired from the storage device with the least access load among the storage devices that store the target data based on the monitoring result The data rebalancing method according to claim 16, wherein the target data is stored in the cache memory, and the target data is transmitted to the information processing terminal. The computer
When there is a request for writing target data from the information processing terminal, based on the monitoring result, the target data is written to the storage device with the least access load among the storage devices, and the target data is stored in the cache memory. Stored in
The data rebalancing method according to claim 16, wherein the target data stored in the cache memory is written to a storage device other than the storage device that performed the writing.

Images